Windflow

Ctrl-World

Stanford University and Tsinghua University · Released October 2025

Ctrl-World is an action-conditioned world model for robot manipulation, trained on the DROID dataset. Its authors roll a vision-language-action policy forward entirely in imagination, so the policy can be evaluated or improved without a physical robot.

Get the weightsCan I use it commercially?
Weights
Open weightsMIT
Commercial use
Permissive license
Real-time
No real-time claim
Where to try
Run it yourself

01 Controls

What you control

  • Keyboard commands converted into robot action chunks
  • Actions proposed by a robot policy

02 License

Can you use it commercially?

Permissive license

The weights' model card declares MIT, a permissive license that allows commercial use. Ctrl-World is fine-tuned from Stable Video Diffusion, whose own license also applies to the base weights it needs.

Declared license: MIT · Model card

A reading of the published license text, not legal advice.

03 Speed

Does it run in real time?

No real-time claim. The cited sources make no real-time claim.

Figures are the authors' own, with their settings and hardware. They are not comparable across models.

04 Setup

How to run it

  1. 01Create a Python 3.11 conda environment and install requirements.txt; to let the π0.5 policy act inside the world model, also install openpi.
  2. 02Download the checkpoint and dataset listed in the README.
  3. 03Replay recorded trajectories, steer the robot arm from the keyboard, or let the π0.5 policy act.

Summarised from the project's own instructions. Follow the source for exact commands and versions.

05 Access

Where to try it

No hosted version was verified. Run the published weights yourself.

Listed alphabetically. Windflow maintains this directory and is one of the hosts; a listing does not promise capacity at any given moment.

06 Questions

Ctrl-World FAQ

Is Ctrl-World open source?

Ctrl-World publishes downloadable weights under MIT. The declared license is permissive.

Can I use Ctrl-World commercially?

The weights' model card declares MIT, a permissive license that allows commercial use. Ctrl-World is fine-tuned from Stable Video Diffusion, whose own license also applies to the base weights it needs.

Does Ctrl-World run in real time?

The cited sources do not claim real-time operation.

Where can I try Ctrl-World?

No hosted version was verified; run the published weights yourself.

What does Ctrl-World take as input?

You give it keyboard commands converted into robot action chunks and actions proposed by a robot policy.

How do I run Ctrl-World?

Following the official instructions: Create a Python 3.11 conda environment and install requirements.txt; to let the π0.5 policy act inside the world model, also install openpi. Download the checkpoint and dataset listed in the README. Replay recorded trajectories, steer the robot arm from the keyboard, or let the π0.5 policy act.

What are alternatives to Ctrl-World?

Similar models in this directory include 1X World Model (open weights, permissive license), EnerVerse-AC (open weights, non-commercial) and ABot-World (open weights, permissive license).

Similar world models

All 56 in the directory